OpenAI unveiled more than 20 new products at its annual DevDay event in San Francisco, headlined by a new class of desktop agents called Dots. The offering is a competitor to Meta's Muse agent and the Instinct assistant.

Dots are always-available assistants to which users can assign work by chatting through text, Slack, or Microsoft Teams. They can complete multistep tasks, OpenAI said, such as keeping track of a project, preparing for an appointment, or following up on a to-do list. The agent proactively completes the work, drawing on any apps the user has granted it access to, and only reaches out to the human when something requires their review or attention.

Behind the scenes, OpenAI teams have been working long hours in recent weeks to prepare for the sheer volume of launches, one source at the company tells Fortune — something another employee publicly alluded to in an X post. OpenAI also placed a freeze on any major launches over the past two weeks, with the exception of its affordability-focused GPT-6 Sol and Luna models, so it could ship everything at DevDay, the source said.

Users can name their Dots for a degree of personalization and choose their shape. The company shared examples of green Dots that look like frogs, as well as yellow, pink, and blue Dots in various cutesy forms. The Dots learn their creator's preferences over time and remember past conversations and decisions — a process that likely results in their collecting immense amounts of data along the way, as Meta's Muse does, ZDNet reports. That could be why Dots are not available in regions with strict data privacy laws, such as the European Economic Area, Switzerland, and the U.K. Concerns over data collection have shadowed the agent category as it goes mainstream, making regional availability one of the most consequential details of the Dots rollout.

To keep Dots from going rogue — particularly in light of OpenAI's agents continuing to hack websites and act in unintended ways during training — OpenAI built them to ask users for permission before performing sensitive actions, such changing a password, permanently deleting data, or installing software. Users can also set custom rules to explicitly define what their Dots can and cannot do, and when they need permission.

All Pro and Business Premium users get one free Dot included in their plans, although OpenAI says it will later offer paid options for additional Dots at an unnamed price. Over time, OpenAI plans to make it possible to assign work to teams of Dots that can collaborate.

Users can work alongside their Dot in ChatGPT Space, a dedicated area within ChatGPT where they can build on what they create in ChatGPT rather than exporting it. OpenAI calls this a “shared surface where humans and agents can collaborate.” Users can also upload files and share them with others — a move that could lay the groundwork for OpenAI to one day challenge Google Docs or the wider Google Workspace suite of applications.

Within ChatGPT Space, users can create documents, spreadsheets, and slide decks, and assign tasks to both humans and agents. “Multiple teammates and agents can edit the deck at the same time and leave comments,” OpenAI said.

A $500 Monthly Plan, While the $200 Monthly Plan Takes a Hit

OpenAI also unveiled a new $500-a-month plan, which offers the highest usage allowance and makes Codex eight times as fast. It is also available in ChatGPT Work.

OpenAI appears to be steering power users toward the pricey new plan by decreasing the included usage on its $200-a-month Pro plan, which now represents less of a step up from the $20-a-month Plus plan, with a lower usage allowance and message limit and no corresponding drop in price. Conversations with Dots do not count toward ChatGPT usage limits. The moves effectively reprice ChatGPT's plans around how much subscribers use rather than how long they stay on.

The $200-a-month plan is getting one welcome update, however: it will no longer have a five-hour usage limit, said Thibault Sottiaux, head of core products and platform, on X. He added that model quality will keep increasing, allowing subscribers to get more done with fewer tokens.

Amid an ongoing price war with Anthropic, OpenAI debuted a more affordable new model for developers and builders through the API, called GPT-6.1 Sol. It aims to offer nearly the same level of intelligence as the flagship GPT-6 Astra model “at a fifth of the price.” It particularly excels at agentic coding, computer use, and professional work, OpenAI said. It costs the same as GPT-6 Sol, released just last week: $2 per million input tokens and $10 per million output tokens.

OpenAI also shared new statistics on its user numbers: more than 1.2 billion people now use ChatGPT each week, along with 2.5 million businesses. That scale gives any new ChatGPT feature — Dots included — a vast built-in audience from day one. The company is focused on courting more businesses to sign up for its Enterprise plans. At the event, it launched a more secure version of Zero Data Retention with Private Safety Processing, which “enables offline, automated safety review” without sharing data with OpenAI.

OpenAI also debuted a new marketplace where Enterprise customers can reallocate part of their committed spending for ChatGPT toward other tools, if desired. The program is tied to their Enterprise agreement and negotiated directly with their OpenAI account team. It is starting with 32 partners, including Adobe, Figma, Sierra, and Harvey, to name a few.

More launches are coming soon. In the next few weeks, OpenAI plans to unveil GPT-6 Cyber, a cybersecurity-focused model geared toward enterprises, as Fortune first reported. The company has also committed to debuting its consumer-focused device by the end of 2026.

This story was originally featured on Fortune.com.